Many rotary power tools, such as drills and screwdrivers, include a clutch for interrupting power output of the tool when the output torque exceeds a threshold value. Generally, such clutches are mechanical or electronic. A mechanical clutch generally includes a spring that biases one clutching member toward another, such that when the torque is exceeded, the clutch members overcome the spring force to separate the clutch members and interrupt torque transmission. One example of a mechanical clutch is shown in U.S. Pat. No. 7,980,324, which is incorporated by reference. However, mechanical clutches tend to wear out over time, are not usually accurately calibrated, and add a great deal of additional length to the tool due to all of the mechanical parts. An electronic clutch generally infers torque indirectly by sensing another parameter of the system, such as current drawn by the motor or motor speed. An example of such an electronic clutch is shown in U.S. Pat. No. 5,410,229, which is incorporated by reference. Such electronic clutches are notoriously inaccurate. First, current is only a good direct indicator of load torque when the motor is rotating at a constant speed, meaning the motor driving duty cycle and load torque has been constant for a significant amount of time. For cases of high motor acceleration and deceleration, as occurs at tool startup or trigger release, current alone is a very poor indicator of load torque. Second, current readings are most often taken through a shunt in series with the motor and battery. This shunt only accurately measures current when the circuit driving the motor is turned on. When the circuit driving the motor is turned off, the shunt experiences no current flow, which creates an artificially low current reading that without compensation is a poor indicator of torque. Third, using current or motor speed to infer torque has a poor response time, as the current or motor speed reading will often lag the actual output torque by a certain time.